Identity as the Attack Surface: Architecting Auth for Autonomous Agents

​The age of sandboxed AI is over. Your agents are now making decisions, delegating tasks, and authenticating across enterprise systems.

​The problem? You're securing dynamic, autonomous agents with static, human-centric IAM tools.

​When an agent spawns a sub-agent to access a database, does it inherit admin privileges? How do you audit a decision made by an ephemeral credential that existed for 30 milliseconds? With machine identities now outnumbering human identities 45 to 1, identity itself has become the attack surface.

​This salon addresses the emerging risks: applying zero-trust principles to non-human identities where LOTL attacks exploit breached agents from within. Securing multi-agent coordination protocols like MCP. Preventing agent impersonation when autonomous systems interact.

​We're gathering the architects solving:

• ​The Delegation Problem: Preventing privilege escalation when agents hand off tasks.

• ​The Ephemeral Stack: Managing just-in-time credentials for high-velocity agent loops.

• ​The New Perimeter: Why identity is the only firewall that matters in an agentic world.